OCR Generated Text Summarization using BART

Optical character recognition (OCR) extracts text from images while models like BART is used for generating summaries and understanding texts. OCR engines transform document images into machine-readable text, but this raw text often contains noise. To enhance text comprehension, we propose an integrated pipeline that combines EasyOCR for text extraction and BART for summarization. EasyOCR is the most recent and most popular OCR method. EasyOCR optically recognizes and extracts text from document images. This extracted text then passed to BART, a transformer-based model designed for sequence-to-sequence tasks including summarization. In BART there are two parts-first part is encoder, which is responsible for understanding text bidirectionally and the second part is decoder which generates the summary in an autoregressive manner. BART is also good for handling noisy data. BART’s token deletion and text infilling tasks are suited to OCR correction, as OCR frequently either misses characters or injects spurious characters.